Everyone is Handing Out Agents

Hi there,

A 23-year-old just raised $1 billion for a personal AI agent. His name is Noah Shinn, he dropped out of college to build an assistant called Instinct, and it's now worth $10 billion, four times what it was a month ago.

Which makes him the second coolest 23-year-old in AI I know 😎

What I'm Seeing in AI

Nobody knows what's going on

The big labs shipped eight frontier models in 30 days. Claude Fable and Mythos 5.1, Gemini 3.8 Flash, Meta's Muse Spark 1.3, Grok 4.7, Claude Opus 5.5, GPT-6 Sol and Luna, Claude Sonnet 5.5, and on Wednesday, Gemini 4 Argon.

That’s a wholeee lot of numbers and words.

Google’s new model from yesterday, came out of nowhere and beat everyone. You can't use it. Only vetted cyber defenders get access, with no public access date.

The same week, OpenAI pulled GPT-6.1 Astra before launch, after it did more than it was asked in testing and was dishonest about what it had done.

CNBC has a name for all this: "model fatigue". CEOs are spending more time comparing models than using them. Sam Altman's answer was "we're all moving to faster cadences", which is not a comforting sentence.

A friend said it best: if you tell me you're an AI expert, no you're not, because it changed this morning. Nobody really knows what's going on. Including me.

Agents, Agents, Agents!

At their conference this week, OpenAI launched Dots: always-on agents inside ChatGPT, each with its own cloud computer and browser, working on your goals around the clock.

Minutes after the keynote, xAI was pushing Grok Bot, which can sit in your team's Slack, connect to your bank and make calls for you. Then there's Instinct, which you text or call like a person.

Some people love these personal agents. Personally, I haven't seen the long-term value yet, and I'd love to be proven wrong.

What I keep noticing is that people love these things for a week and then quietly stop using them. Lots of people have a cool agent story but fewer still open it.

Inside our own company, the always-on, chat-with-it agent is the thing we use least. What actually stuck is automations that run behind the scenes: email triaged before I open it, or a skill that does the boring part of a job. I think there's a simple reason. Nobody wants a new colleague to talk to. They want the work to be done.

Maybe Dots changes that. We shall see.

What I'm Building

Our marketing interns are COOKING

Last week I said our distribution needed to catch up with the product. Matt and Frank started, and in week one they shipped:

- Our first customer film, on how a Telegram agent saves a recruiter 5-10 hours a week (watch it here)

- thehourglass.ai/results, one page per client result, with every number checked

- A YouTube channel, every social bio rewritten, and a reel with Jeremy

- A skill (Frank's) that drafts a newsletter from our meetings and Slack, then checks its own work before anyone reads it

Our first customer film. Tap to watch.

Videos with Anna on evals are coming out this week.

Neither of them is an engineer.

We're hiring: the unicorn skill

We're hiring forward-deployed engineers in Melbourne & Sydney. That means sitting with a client, understanding how they actually work, then building the AI to do the work.

The role started at Palantir, which until 2016 had more of them than regular software engineers. Now OpenAI, Anthropic and Salesforce are all hiring them, and job postings are up 729% in a year.

Where the forward-deployed engineer sits (via The Pragmatic Engineer).

It's rare because you need both halves. You have to be good with people, able to sit in a room and work out what the real problem is. And you have to be genuinely good at building with AI. Most people are one or the other.

If that's you or you know someone, please send a short video at thehourglass.ai/hiring.

Know someone who's great with people and great with AI? Send them my way.

And if you're using one of these personal agents every day, reply too. I want to hear what I'm missing.

If this was useful, forward it to someone trying to keep up.

See you next Friday,

Finlay

Explore the rest of my writing here → finlayekins.com/writing